Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Snowy Scenes,Clear Detections: A Robust Model for Traffic Light Detection in Adverse Weather Conditions

Shivank Garg, Abhishek Baghel, Amit Agarwal, Durga Toshniwal

TL;DR

Problem: Robust traffic-light detection under adverse weather is hindered by domain shifts between snow-affected test data and standard training data. Approach: A pipeline that generates synthetic snowy imagery from ground-truth data and blends it with real data, followed by fine-tuning across Detectron2/Faster R-CNN and YOLO v7/v8 detectors. Contributions: creation of a snow-augmented dataset, extensive evaluation across models, and demonstrated gains in IoU, mAP, and F1, including substantial improvements under domain shifts. Findings: the method yields about 40.8% improvement in average IoU and F1 over naive fine-tuning and about 22.4% under domain-shift scenarios. Significance: improves reliability and safety for autonomous driving in snow, fog, rain, and related adverse weather.

Abstract

With the rise of autonomous vehicles and advanced driver-assistance systems (ADAS), ensuring reliable object detection in all weather conditions is crucial for safety and efficiency. Adverse weather like snow, rain, and fog presents major challenges for current detection systems, often resulting in failures and potential safety risks. This paper introduces a novel framework and pipeline designed to improve object detection under such conditions, focusing on traffic signal detection where traditional methods often fail due to domain shifts caused by adverse weather. We provide a comprehensive analysis of the limitations of existing techniques. Our proposed pipeline significantly enhances detection accuracy in snow, rain, and fog. Results show a 40.8% improvement in average IoU and F1 scores compared to naive fine-tuning and a 22.4% performance increase in domain shift scenarios, such as training on artificial snow and testing on rain images.

Snowy Scenes,Clear Detections: A Robust Model for Traffic Light Detection in Adverse Weather Conditions

TL;DR

Problem: Robust traffic-light detection under adverse weather is hindered by domain shifts between snow-affected test data and standard training data. Approach: A pipeline that generates synthetic snowy imagery from ground-truth data and blends it with real data, followed by fine-tuning across Detectron2/Faster R-CNN and YOLO v7/v8 detectors. Contributions: creation of a snow-augmented dataset, extensive evaluation across models, and demonstrated gains in IoU, mAP, and F1, including substantial improvements under domain shifts. Findings: the method yields about 40.8% improvement in average IoU and F1 over naive fine-tuning and about 22.4% under domain-shift scenarios. Significance: improves reliability and safety for autonomous driving in snow, fog, rain, and related adverse weather.

Abstract

With the rise of autonomous vehicles and advanced driver-assistance systems (ADAS), ensuring reliable object detection in all weather conditions is crucial for safety and efficiency. Adverse weather like snow, rain, and fog presents major challenges for current detection systems, often resulting in failures and potential safety risks. This paper introduces a novel framework and pipeline designed to improve object detection under such conditions, focusing on traffic signal detection where traditional methods often fail due to domain shifts caused by adverse weather. We provide a comprehensive analysis of the limitations of existing techniques. Our proposed pipeline significantly enhances detection accuracy in snow, rain, and fog. Results show a 40.8% improvement in average IoU and F1 scores compared to naive fine-tuning and a 22.4% performance increase in domain shift scenarios, such as training on artificial snow and testing on rain images.
Paper Structure (17 sections, 3 figures, 2 tables)

This paper contains 17 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Object Detection during Bad weather condition:
  4. Recent Technique for object detection:
  5. Faster RCNN
  6. YOLO
  7. YOLO V7
  8. YOLO V8
  9. Method and Experiments
  10. Dataset
  11. Synthetic Dataset Generation
  12. Fine Tuning
  13. Results
  14. Validation Results
  15. Results Based on Domain Shift and Real Examples
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Our Pipeline: We create a synthetic version of the ground truth image and use either the original or the synthesized one for training the model.
  • Figure 2: Comparison of YOLO v8 tested for naturally occurring real images. Left: Images tested on model fine-tuned on the proposed pipeline for artificial snow dataset. Right: Images tested on model model fine-tuned on ground truth traffic light dataset
  • Figure 3: Visual Comparison of Results on Validation Set Obtained by YOLO v8 Model